Confidence-Guided Pseudo-Label Refinement for Unsupervised Cross-Domain Person Re-Identification

Key points

• Unsupervised cross-domain person re-identification remains challenging because domain shifts generate noisy pseudo labels that are reinforced during iterative adaptation.1
• Existing clustering-based methods often rely on limited reliability cues or global confidence criteria and may impose hard identity supervision on ambiguous target samples.1
• CGPLR estimates sample-level reliability by jointly modeling centroid compatibility, local neighborhood consistency, separation from the strongest competing cluster and temporal assignment stability.1
• The learning objective adapts to these decisions through confidence-weighted supervision for preserved samples, cautious corrected-label weighting for reassigned samples and label-free representation consistency for deferred samples.1

Sources (1)

• [1]Confidence-Guided Pseudo-Label Refinement for Unsupervised Cross-Domain Person Re-Identification
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 19, 01:24 AM

  To address these limitations, we propose CGPLR, a confidence-guided pseudo-label refinement framework for unsupervised domain-adaptive person re-identification.

  Unsupervised cross-domain person re-identification remains challenging because domain shifts generate noisy pseudo labels that are reinforced during iterative adaptation.

Extractive summary: sentences quoted from the sources.